cond-mat0506629/seb.tex
1: \documentclass[10pt]{iopart}
2: \usepackage{epsfig}
3: \newcommand{\be}{\begin{equation}}
4: \newcommand{\ee}{\end{equation}}
5: \newcommand{\tw}{t_{\rm w}}
6: \newcommand{\ta}{\tau_\alpha}
7: \newcommand{\tdh}{\tau_{\rm dh}}
8: \begin{document}
9: 
10: \title[Lifetime of dynamic heterogeneity 
11: in spin facilitated models]
12: {Lifetime of dynamic heterogeneity in strong and 
13: fragile kinetically constrained spin models}
14: 
15: \author{S\'ebastien L\'eonard and Ludovic 
16: Berthier\footnote[2]{To whom correspondence should be addressed:
17: berthier@lcvn.univ-montp2.fr}}
18: 
19: \address{Laboratoire des Collo\"{\i}des, Verres et Nanomat\'eriaux,
20: UMR 5587 CNRS and  Universit\'e Montpellier II,
21: 34095 Montpellier Cedex 5, France}
22: 
23: \begin{abstract}
24: Kinetically constrained spin models are schematic
25: coarse-grained models for the glass transition 
26: which represent an efficient theoretical tool to study 
27: detailed spatio-temporal aspects of dynamic
28: heterogeneity in supercooled liquids.
29: Here, we study how spatially correlated 
30: dynamic domains evolve with time and compare our results
31: to various experimental and numerical investigations.
32: We find that strong and fragile models yield different 
33: results. In particular, the lifetime of dynamic heterogeneity 
34: remains constant and roughly equal 
35: to the alpha relaxation time in strong models, while 
36: it increases more rapidly in fragile models
37: when the glass transition is approached.
38: \end{abstract}
39: 
40: \section{Introduction}
41: \label{Intro}
42: 
43: The viscosity of supercooled liquids increases
44: extremely rapidly when the temperature is reduced towards
45: the glass temperature. It is firmly established that 
46: this dramatic slowing down is spatially heterogeneous.
47: Local relaxation is widely distributed
48: in time --- existence of broad stretched relaxations ---, but
49: also in space --- existence of dynamic heterogeneity~\cite{ediger}.
50: The main physical aspect is that spatial 
51: fluctuations of local relaxations become increasingly spatially
52: correlated when temperature decreases. Direct
53: experimental investigations of the time and temperatures 
54: dependences of the relevant dynamic lengthscales at low temperature 
55: are however still missing.
56: 
57: To study dynamic heterogeneity, statistical correlators which probe
58: more than two points in space and time have to be 
59: considered~\cite{ediger,sillescu,glotzer}.
60: For example, if one wants to study spatial correlations
61: of the local dynamics one has to define a two-point, 
62: two-time correlator, 
63: \be
64: \label{22}
65: C_{2,2}(|i-j|,t) = \langle P_i(0, t) P_j(0, t) \rangle
66: -\langle P_i(0, t) \rangle \langle P_j(0, t) \rangle,
67: \ee  
68: where notations are adapted to 
69: lattice spin models. In equation~(\ref{22}), $(i, j)$ denote 
70: lattice sites, $P_i(0, t)$ quantifies
71: the dynamics at site $i$ between times 0 and $t$ (autocorrelation
72: or persistence functions), and brackets represent ensemble averages.
73: The physical meaning of (\ref{22}) is clear: given a spontaneous 
74: fluctuation of the two-time dynamics at site $i$, is there a similar 
75: fluctuation at site $j$? The quantity (\ref{22}) has now been 
76: discussed both theoretically and numerically in some 
77: detail~\cite{glotzer,glotzer2,wyart}, 
78: generically revealing the existence of a growing spatial range 
79: of dynamic correlations in supercooled liquids accompanying 
80: an increasingly sluggish dynamics.
81: 
82: Logically, the next question is then: given spatial structures
83: of the local 
84: relaxation between times 0 and $t$, what will this structure look like,
85: say, between times $t$ and $2t$? In other words~\cite{heuer1}, 
86: how do dynamic heterogeneities evolve with time?
87: This question is in fact simpler to address experimentally
88: because no spatial resolution is needed 
89: and different experimental techniques can be devised:
90: NMR~\cite{nmr}, solvation dynamics~\cite{solvation}, 
91: optical~\cite{ediger3} and dielectric~\cite{hole} hole-burning.
92: In statistical terms, one wants to study
93: a four-time correlation function of the general form
94: \be
95: C_4(t_1, \tw, t_2) = \langle P_i(0, t_1) P_i(t_1+\tw, t_1+\tw+t_2) \rangle, 
96: \label{4}
97: \ee
98: which correlates dynamics between times 0 and $t_1$
99: and between times $t_1+\tw$ and $t_1+\tw+t_2$. Again the 
100: physical content of (\ref{4}) is clear~\cite{heuer2}: given a dynamic
101: fluctuation at site $i$ between in a certain time interval $t_1$, 
102: how long does it take for this fluctuation to be washed out? 
103: This leads to the general concept of a lifetime, $\tdh$, for dynamic
104: heterogeneity. While many 
105: investigations~\cite{ediger,sillescu,nmr,solvation,hole,harrowell,szamel} 
106: indicate that $\tdh$
107: is in fact slaved to the alpha-relaxation time of the 
108: liquid, $\tdh \approx \ta$, photobleaching experiments 
109: very close to the glass transition
110: indicate  that $\tdh$ may become several 
111: orders of magnitude larger than $\ta$, although in a
112: surprisingly abrupt manner~\cite{ediger2}. 
113: 
114: In this paper we study the lifetime of dynamic heterogeneity
115: in kinetically constrained spin models of supercooled 
116: liquids~\cite{reviewkcm}. 
117: These models represent schematic coarse-grained models 
118: for the glass transition and provide a very efficient tool to study 
119: in detail many spatio-temporal aspects related to dynamic
120: heterogeneity such as  dynamic lengthscales~\cite{gc}, 
121: scaling~\cite{steve}, or decoupling 
122: phenomena~\cite{jung,epl}. 
123: They are simple enough that analytical 
124: progress can be made and numerical simulations performed 
125: on a wide range of lengthscales and timescales, and yet 
126: rich enough that direct comparisons to both simulations
127: and experiments can be made.
128: 
129: \section{Models} 
130: 
131: Following previous works~\cite{gc,jung,epl,BG}, 
132: we focus on two specific 
133: spin facilitated models in one spatial dimension, namely 
134: the one-spin facilitated 
135: Fredrickson-Andersen (FA) model~\cite{fa}
136: and the East model~\cite{east} that respectively behave as strong 
137: and fragile systems~\cite{reviewkcm}. 
138: These are probably 
139: the simplest models which  incorporate the ideas that (i) mobility
140: in supercooled liquids is both highly localized and sparse, as
141: revealed by simulations~\cite{harrowell}; 
142: (ii) a localized mobility very 
143: easily propagates to neighbouring regions, the 
144: dynamic facilitation concept. 
145: Detailed studies in spatial dimensions larger than one have shown
146: that dimensionality does not play a relevant qualitative 
147: role~\cite{steve,jung,nef},
148: and justify therefore the present one-dimensional studies. 
149: 
150: Both models are defined by the same non-interacting Hamiltonian, 
151: $H = \sum_i n_i$, expressed in terms of a mobility variable, $n_i=1$
152: when site $i$ is mobile, $n_i=0$ otherwise.
153: Dynamic facilitation is incorporated at the level of the 
154: dynamic rules through kinetic constraints. 
155: In the FA model, the site $i$ can evolve with Boltzmann 
156: probability if at least one of its two neighbours is mobile, 
157: $n_{i-1}+n_{i+1} > 0$. 
158: In the East model the site $i$ can evolve only if its left 
159: neigbour is mobile, $n_{i-1}=1$.
160: 
161: We have performed numerical simulations of both models
162: using a continuous time Monte Carlo algorithm where all moves
163: are accepted and the time is updated according to the 
164: corresponding statistical 
165: weight. Simulations have been performed 
166: `only' over about 7 decades in time because 
167: extensive time averaging is required to accurately 
168: measure multi-time correlation functions such as equation~(\ref{4}).
169: 
170: \section{Results}
171: 
172: \subsection{Dynamic filtering}
173: 
174: There are several parameters involved in the four-time 
175: correlator (\ref{4}) that need to be appropriately chosen.
176: Since dynamic heterogeneity is more pronounced for times 
177: close to $\ta$ it is sensible to first fix 
178: $t_1 = \ta$ and to study the remaining $\tw$ 
179: and $t_2$ dependences.
180: As a local dynamic correlator we first
181: focus on the persistence function~\cite{heuer5}, 
182: $P_i(0,t)=1$ if spin $i$ has not flipped in the interval $[0,t]$, 
183: $P_i(0,t)=0$ otherwise. 
184: We also define the mean persistence, 
185: $p(t) = \langle P_i(t) \rangle$, from which we measure 
186: $\ta$ via $p(\ta)=e^{-1}$.   
187: 
188: \begin{figure}
189: \begin{center}
190: \psfig{file=fig1a.ps,height=6.45cm,angle=-90}
191: \psfig{file=fig1b.ps,height=6.45cm,angle=-90}
192: \caption{\label{fig1}  Left: Four-time correlation
193: function $C_4(\tau_\alpha, \tw, t_2)$ normalized by its 
194: $t_2=0$ value for various $\tw=0$, 
195: 550, 2100, 5200, 20000, 49500, and 191000 (from right to left)
196: in the FA model at $T=0.3$. These dashed curves converge 
197: at large $\tw$ to the bulk persistence function, $p(t_2)$,  
198: shown as a full line.
199: Right: Corresponding logarithmic distributions of relaxation 
200: times.}
201: \end{center}
202: \end{figure}
203: 
204: In \fref{fig1} (left) we show the $t_2$ dependence
205: of $C_4(\tau_\alpha, \tw, t_2)$ for various $\tw$
206: at $T=0.3$ in the FA model. We have normalized 
207: $C_4$ by $p(\ta)=e^{-1}$, its value at $t_2=0$.
208: By definition, this function describes the persistence 
209: function in the interval $[\tw+\ta, \tw+\ta+t_2]$ of those sites
210: which had not flipped in the interval $[0, \ta]$, 
211: and were therefore slower than 
212: average.
213: The first term in the correlator (\ref{4}) plays the role
214: of a dynamic filter~\cite{heuer2}, selecting a sub-population of sites
215: which have an average dynamics different from the bulk.
216: From earlier works studying the spatial correlator
217: (\ref{22}), it is known that those sites belong to 
218: compact clusters that represent the largest regions of space with 
219: no mobility defects at time 0~\cite{gc,steve,nef}.
220:  
221: Immediately after filtering one expects therefore those slow regions
222: to remain slow, as 
223: indeed observed in \fref{fig1} for $\tw=0$. 
224: When $\tw$ increases, this selected population gradually forgets
225: it was initially slow. When $\tw \to \infty$, bulk
226: dynamics is recovered,
227: \be 
228: \frac{C_4(t_1, \tw \to \infty, t_2)}{p(\ta)} \to p(t_2), 
229: \ee
230: as demonstrated by the full line in \fref{fig1}. 
231: In \fref{fig1} (right) we also show the 
232: (logarithmic) distribution 
233: of relaxation times corresponding to the functions shown 
234: in the left panel, a representation sometimes preferred 
235: in experimental works~\cite{ediger}. Both quantities are of course
236: fully equivalent~\cite{BG}. 
237: It is clear from these figures that once a subset 
238: of sites has been dynamically selected the remaining relaxation
239: is narrower than the bulk relaxation. In fact all persistence functions
240: shown in \fref{fig1} are well described by stretched 
241: exponentials. While $\beta=1/2$ is observed for the bulk dynamics, 
242: one finds $\beta \approx 0.83$ at $\tw=0$. Accordingly, distribution
243: of relaxation times progressively broaden 
244: when $\tw$ increases. These results are consistent with experimental 
245: observations.
246: 
247: In the FA model their 
248: interpretation is straightforward. Stretching in this model
249: follows from an exponential distribution of distance between mobility 
250: defects~\cite{gc}.
251: Dynamic filtering implies that this domain distribution is cut-off 
252: at small distance. Narrower lengthscale distributions 
253: directly imply narrower timescale distributions. 
254: 
255: We have also investigated the effect of changing the 
256: `filter efficiency'~\cite{nmr} which
257: in our case implies changing the duration of the filtering interval, 
258: $[0,t_1]$. While bulk distributions are found for 
259: $t_1/\ta \ll 1$ (weak filtering) 
260: distributions shift to larger times and become very narrow 
261: when $t_1/\ta$ increases. 
262: In the following we work at constant 
263: filtering, $t_1 = \ta$. 
264: 
265: \subsection{`Homogeneous' vs. `heterogeneous' dynamics}
266: 
267: The ability to select a sub-ensemble of sites that 
268: are slower than average is sometimes taken as a definition of 
269: dynamic heterogeneity~\cite{nmr}, although lengthscales play no role
270: in this view. That FA and East model display 
271: spatially heterogeneous dynamics is well-known, 
272: and the results of the previous section are therefore 
273: natural. 
274: 
275: \begin{figure}
276: \begin{center}
277: \psfig{file=fig3a.ps,height=6.4cm,angle=-90}
278: \psfig{file=fig3b.ps,height=6.4cm,angle=-90}
279: \caption{\label{fig3} $F_3(k,t)$ and its `heterogeneous', $F_s(k,t)$,
280: and `homogeneous', $[F_s(k,t/2)]^2$, limits.
281: Left: East model for fixed $k=\pi/5$ and various temperatures. 
282: Right: East model at $T=0.4$ and various wavevectors.}  
283: \end{center}
284: \end{figure}
285: 
286: Another indicator of dynamic heterogeneity has been 
287: proposed~\cite{heuer2,heuer3,heuer4} 
288: based on the analysis of the four-time correlation
289: (\ref{4}). Consider the situation where $\tw=0$, and $t_1=t_2 \equiv t/2$. 
290: In that case, one studies a `three-time' correlation~\cite{heuer4}
291: \be 
292: F_3(t) = \langle P_i(0,t/2) P_i(t/2,t) \rangle.
293: \label{F4}
294: \ee 
295: Two extreme behaviours can be expected for $F_3(t)$. (i) Dynamics 
296: in the intervals $[0,t/2]$ and $[t/2,t]$ are totally uncorrelated, 
297: and thus $F_3(t) \approx [p(t/2)]^2$.
298: (ii) Dynamics in the two intervals are strongly correlated, in the sense 
299: that those regions that survive 
300: filtering in $[0,t/2]$ are also those dominating the relaxation
301: in the full interval $[0,t]$. In that case, $F_3(t) \approx p(t)$. 
302: Scenarii (i) and (ii) have been termed `homogeneous' and 
303: `heterogeneous', respectively, although again lengthscales play no
304: role in the distinction.
305: Clearly, both estimates become equivalent
306: when $p(t)$ decays exponentially.
307: 
308: Of course when studying the persistence function in the FA and East models, 
309: scenario (ii) strictly applies by definition, because 
310: $P_i(0,t/2) P_i(t/2,t) = P_i(0,t)$. In real materials,
311: smoother dynamic functions are studied, directly defined 
312: from the particles positions instead of a mobility field. 
313: Our strategy is therefore to couple probe particles to our 
314: mobility field, see Refs.~\cite{jung,epl} for technical details. 
315: From probe molecule displacements, $\delta x(0,t) = x(t)-x(0)$, 
316: we define self-intermediate scattering functions, $F_s(k,t) = 
317: \langle \cos[k \cdot \delta x (0,t) ] \rangle$, and the analog of 
318: equation (\ref{F4}), $F_3(k,t) = \langle \cos [k \cdot \delta x (0,t/2)] 
319: \cos [k \cdot \delta x (t/2,t)] \rangle$. 
320: 
321: Our numerical results are presented 
322: in \fref{fig3}.
323: Clearly the time dependence of $F_3$ closely 
324: follows the one of $F_s(k,t)$, in agreement with the `heterogeneous' scenario
325: described above. This is consistent with numerical
326: results~\cite{heuer4}. 
327: 
328: In the present approach, this result is a natural consequence
329: of decoupling between structural relaxation and 
330: diffusion~\cite{jung,epl}.
331: At large wavevectors, $k \sim \pi$, 
332: corresponding to distances of the order of the lattice 
333: spacing, $F_s(k,t)$ is dominated by the time distribution 
334: of the first jump of the probe molecule in the interval $[0,t]$, 
335: so that $F_s(k,t) \approx p(t) \approx F_3(k,t)$.
336: At large distance, $k < k^*$, Fickian diffusion holds~\cite{epl},
337: $F_s(k,t) = \exp(-k^2 D_s t)$, and there is no distinction between
338: homogeneous and heterogeneous relaxation.
339: At intermediate wavectors, $\pi > k > k^*$, 
340: the long-time decay of $F_s(k,t)$ is again
341: dominated by the persistence time distribution,
342: because the timescale it takes a molecule to make
343: $2 \pi / k$ steps is strongly dominated by the 
344: timescale to make the first step~\cite{epl}.
345: This is just the condition for the heterogeneous 
346: scenario to hold, in agreement with \fref{fig3}.
347: The characteristic wavevector separating the two regimes,
348: $k^*(T) = 1/\sqrt{\ta D_s}$, decreases when temperature decreases, 
349: opening a larger heterogeneous window; $k^*$ also 
350: sets the upper limit of validity of Fickian diffusion
351: in supercooled liquids~\cite{epl}.
352: 
353: \subsection{Lifetime of dynamic heterogeneity}
354: 
355: After dynamic filtering it takes some time for filtered
356: distributions to reequilibrate towards the bulk relaxation, 
357: cf \fref{fig1}.
358: To extract the typical lifetime of dynamic heterogeneity, $\tdh$, 
359: we tried several procedures which all lead to similar results, based
360: on how timescales (time decay of persistence functions
361: or moments of the corresponding distributions)
362: return to their equilibrium values.
363: Following Refs.~\cite{szamel,jung} we also measured the
364: integrated difference between filtered and bulk dynamics, 
365: $\Delta (\tw) \equiv \int_0^\infty dt_2 [
366: C_4(\ta, \tw, t_2)/p(\ta) - p(t_2)]$.
367: From \fref{fig1}, we expect that $\Delta (\tw)$ goes to 0
368: on a timescale $\tdh$. In practice, we define
369: $\tdh$ as $\Delta(\tdh)/\Delta(0) = e^{-1}$. 
370: In principle, $\tdh$ depends on the filtering time, $t_1$, and on
371: temperature, $T$. 
372: 
373: \begin{figure}
374: \begin{center}
375: \psfig{file=fig4a.ps,height=6.4cm,angle=-90}
376: \psfig{file=fig4b.ps,height=6.05cm,angle=-90}
377: \caption{\label{fig4} Integrated difference between 
378: filtered and bulk dynamics at constant filtering time, $t_1 = \ta$, 
379: and various temperatures in the FA (left) and East (right) models. 
380: Times have been rescaled to collapse the data points at various 
381: temperatures and extract the lifetime of dynamic heterogeneity, $\tdh$. 
382: In the FA model, $\tdh$ is set by $\ta$ with tiny corrections
383: that vanish at low $T$. In the East model, 
384: $\tdh$ grows faster than $\ta$, as emphasized in the inset
385: showing the systematic increase of $\tdh/\ta$ as $T$ decreases.
386: Mastercurves are fitted by stretched exponentials shown 
387: with full lines, $\beta=0.75$ (left) and $\beta=0.8$ (right).}  
388: \end{center}
389: \end{figure}
390: 
391: We show in \fref{fig4} results at 
392: various $T$ but constant 
393: filter efficiency, $t_1 = \tau_\alpha(T)$, in East 
394: and FA models. 
395: While $\tdh$ is set by $\ta$ in the FA model 
396: (the tiny deviation observed in \fref{fig4} 
397: is due to finite $T$ corrections which weaken when $T$ 
398: gets lower),
399: this is not true in the East model where $\tdh$ 
400: systematically grows faster than 
401: $\ta$ at low $T$, as emphasized in the inset.
402: Quantitatively a power law relationship, 
403: $\tdh \sim \ta^{1+\zeta}$, with $\zeta \approx 0.06$,  
404: is a good description of the data, although alternative 
405: fitting formula could probably be used. 
406: 
407: In the fragile case, $\tdh$ can therefore be considered as an additional 
408: slow timescale characterizing the 
409: alpha-relaxation~\cite{ediger2}, on top 
410: of $\ta$ and $1/D_s$~\cite{jung}. 
411: The comparative study of FA and East models
412: offers a possible physical interpretation. While both models
413: display stretched relaxations, in the FA model stretching is constant,
414: $\beta=1/2$, while $\beta$ increases linearly with $T$ in the East 
415: model~\cite{BG}. Therefore $\ta$ represents the first moment
416: of a distribution that becomes wider and wider 
417: when $T$ decreases. We attribute the small but systematic 
418: decoupling between $\tdh$ and $\ta$ to this broadening. 
419: 
420: Unfortunately this decoupling does not quantitatively account for 
421: the results of photobleaching experiments which show that 
422: $\tdh/\ta$ increases strongly close to $T_g$~\cite{ediger2}. 
423: In OTP, while 
424: $\ta$ changes by about 1 decade when $T$ is changed from $T_g+4$~K to
425: $T_g+1$~K, the ratio $\tdh/\ta$ changes by 2 orders of magnitude, 
426: so that $\zeta \approx 2$. This value is much too large 
427: to be accounted for by the above results. Presumably, also, $\beta$ does
428: not vary much on such a tiny temperature interval. 
429: Therefore, the present results cannot explain 
430: the experimental value $\zeta \approx 2$ without invoking 
431: possible non-equilibrium effects due to the proximity of 
432: $T_g$. However we were able to predict instead 
433: a smaller, but definitely non-vanishing decoupling between 
434: the lifetime of dynamic heterogeneity and 
435: the alpha-relaxation time which could be detected in 
436: dynamic filtering experiments performed on a sufficiently large temperature
437: window in fragile glass-formers.  
438: 
439: \Bibliography{99}
440: 
441: \bibitem{ediger} M.D. Ediger, Annu. Rev. Phys. Chem. {\bf 51}, 99 (2000).
442: 
443: \bibitem{sillescu} H. Sillescu, J. Non-Crystalline Solids {\bf 243},
444: 81 (1999).
445: 
446: \bibitem{glotzer} S.C. Glotzer, J. Non-Crystalline Solids {\bf 274},
447: 342 (2000).
448: 
449: \bibitem{glotzer2} C. Bennemann, C. Donati, J. Baschnagel, and 
450: S.C. Glotzer, Nature {\bf 399}, 246 (1999);
451: N. Lacevic, F.W. Starr, T.B. Schr{\o}der, and S.C. Glotzer,
452: J. Chem. Phys. {\bf 199}, 7372 (2003).  
453: 
454: \bibitem{wyart} C. Toninelli, M. Wyart, L. Berthier, G. Biroli, 
455: and J.-P. Bouchaud,
456: Phys. Rev. E {\bf 71}, 041505 (2005). 
457: 
458: \bibitem{heuer1} B. Doliwa and A. Heuer, 
459: J. Non-Crystalline Solids {\bf 307-310}, 32 (2002).
460: 
461: \bibitem{nmr} 
462: K. Schmidt-Rohr and H.W. Spiess, Phys. Rev. Lett. {\bf 66},
463: 3020 (1992);
464: R. B\"ohmer, G. Diezemann, G. Hinze, and H. Sillescu,
465: J. Chem. Phys. {\bf 108}, 890 (1998). 
466: 
467: \bibitem{solvation} M. Yang and R. Richert, 
468: J. Chem. Phys. {\bf 115}, 2676 (2001).
469: 
470: \bibitem{ediger3} M.D. Ediger, 
471: J. Chem. Phys. {\bf 103}, 752 (1995).
472: 
473: \bibitem{hole} B. Schiener, R. B\"ohmer, A. Loidl, and
474: R.V. Chamberlin, Science {\bf 274}, 752 (1996).
475: 
476: \bibitem{heuer2} A. Heuer, Phys. Rev. E {\bf 56}, 730 (1997).
477: 
478: \bibitem{harrowell} D.N. Perera and P. Harrowell, 
479: J. Chem. Phys. {\bf 111}, 5441 (1999).
480: 
481: \bibitem{szamel} E. Flenner and G. Szamel, 
482: Phys. Rev. E {\bf 70}, 052501 (2004). 
483: 
484: \bibitem{ediger2} C.-Y. Wang and M.D. Ediger, J. Phys. Chem. B {\bf 103},
485: 4177 (1999).
486: 
487: \bibitem{reviewkcm} F. Ritort and P. Sollich, 
488: Adv. Phys. {\bf 52}, 219 (2003). 
489: 
490: \bibitem{gc} J.P. Garrahan and D. Chandler, 
491: Phys. Rev. Lett. {\bf 89}, 035704 (2002). 
492: 
493: \bibitem{steve} S. Whitelam, L. Berthier, and J.P. Garrahan, 
494: Phys. Rev. Lett. {\bf 92}, 185705 (2004); Phys. Rev. E {\bf 71}, 
495: 026128 (2005).
496: 
497: \bibitem{jung} Y. Jung, J.P. Garrahan and D. Chandler, 
498: Phys. Rev. {\bf 69}, 061205 (2004); preprint cond-mat/0504535.
499: 
500: \bibitem{epl} L. Berthier, D. Chandler, and J.P. Garrahan,
501: Europhys. Lett. {\bf 69}, 230 (2005).
502: 
503: \bibitem{BG} L. Berthier and J.P. Garrahan, 
504: J. Chem. Phys. {\bf 119}, 4367 (2003); 
505: Phys. Rev. E {\bf 68}, 041201 (2003).
506: 
507: \bibitem{fa} G.H. Fredrickon and H.C. Andersen, 
508: Phys. Rev. {\bf 53}, 1244 (1984). 
509: 
510: \bibitem{east} J. J\"ackle and S. Eisinger, 
511: Z. Phys. B {\bf 84}, 115 (1991).
512: 
513: \bibitem{nef} L. Berthier and J.P. Garrahan, 
514: J. Phys. Chem. {\bf 109}, 3578 (2005).
515: 
516: \bibitem{heuer5} A. Heuer, U. Tracht, and H.W. Spiess,
517: J. Chem. Phys. {\bf 107}, 3813 (1997). 
518: 
519: \bibitem{heuer3} A. Heuer and K. Okun, 
520: J. Chem. Phys. {\bf 106}, 6176 (1997).
521: 
522: \bibitem{heuer4} B. Doliwa and A. Heuer, Phys. Rev. Lett. {\bf 80}, 
523: 4915 (1998).
524: 
525: \endbib
526: 
527: \end{document}
528: